thomas, jennifer; massachusetts general hospital, eating

Acc

epte

d A

rtic

leThomas, Jennifer; Massachusetts General Hospital, Eating Disorders Clinical & Research Program; Harvard Medical School, Psychiatry

Keywords: eating disorder, meta-analysis, treatment outcome, therapeutic alliance

of 54 International Journal of Eating Disorders

This is the author manuscript accepted for publication and has undergone full peer review but has not beenthrough the copyediting, typesetting, pagination and proofreading process, which may lead to differencesbetween this version and the Version record. Please cite this article as doi:10.1002/eat.22672.

This article is protected by copyright. All rights reserved.

http://dx.doi.org/10.1002/eat.22672

http://dx.doi.org/10.1002/eat.22672

Acc

epte

d A

rtic

leRunning head: META-ANALYSIS OF THE RELATION BETWEEN THERAPEUTIC

ALLIANCE 1

A Meta-Analysis of the Relation between Therapeutic Alliance and Treatment Outcome in

Eating Disorders

Tiffany A. Gravesa, Nassim Tabri

a, Heather Thompson-Brenner

a, Debra L. Franko

a, and

Kamryn T. Eddya

aMassachusetts General Hospital and Harvard Medical School

Stephanie Bourion-Bedesb

bCentre Hospitalier Lorquin

Amy Brownc

cEating Disorder Service, South London and Maudsley NHS Foundation Trust

Michael J. Constantinod

dUniversity of Massachusetts-Amherst

Christoph Flückigere

eUniversity of Wisconsin-Madison and University of Bern

Sarah Forsbergf

fStanford University School of Medicine

Leanna Isserling

gUniversity of Ottawa

Jennifer Couturierh

hMcMaster Children’s Hospital

Gunilla Paulson Karlssoni

iThe Sahlgrenska University Hospital

Johannes Manderj

jUniversity of Heidelberg

of 54

International Journal of Eating Disorders



Acc

epte

d A

rtic

leMETA-ANALYSIS OF THE RELATION BETWEEN THERAPEUTIC ALLIANCE

2

Martin Teufelk

kMedical University Hospital

James E. Mitchelll

lUniversity of North Dakota School of Medicine and Health Sciences and the Neuropsychiatric

Ross D. Crosbym

mNeuropsychiatric Research Institute and University of North Dakota School of Medicine and

Health Sciences

Claudia Prestanon

nNiccolò Cusano Roma

Dana A. Satiro

oUniversity of Denver

Susan Simpsonp

pUniversity of South Australia

Richard Slyq

qUniversity of East Anglia

J. Hubert Laceyr

rSt. George's University of London

Colleen Stiles-Shieldss

sNorthwestern University Feinberg School of Medicine and University of Chicago

Giorgio A. Tascat

tUniversity of Ottawa and the Ottawa Hospital

Glenn Walleru

uUniversity of Sheffield

of 54




Acc

epte

d A

rtic


3

Shannon L. Zaitsoffv

vSimon Fraser University

Renee Rieneckew

wThe University of Michigan Comprehensive Eating Disorders Program

Daniel Le Grangex

xUniversity of California, San Francisco

Jennifer J. Thomasy

yMassachusetts General Hospital and Harvard Medical School

Correspondence concerning this article should be addressed to Jennifer J. Thomas, Ph.D., Eating

Disorders Clinical and Research Program, Massachusetts General Hospital, 2 Longfellow Place,

Suite 200, Boston, MA 02114. Phone: 617-643-6306. Fax: 617-726- 1595. Email:

[email protected]

of 54




Acc

epte

d A

rtic


4

Abstract

Objective: The therapeutic alliance between patient and therapist has demonstrated an

association with favorable psychotherapeutic outcomes in the treatment of eating disorders

(EDs). However, questions remain about the inter-relationships between early alliance, early

symptom improvement, and treatment outcome. We conducted a meta-analysis on the relations

among these constructs, and possible moderators of these relations, in psychosocial treatments

for EDs. Method: Twenty studies met inclusion criteria and supplied sufficient supplementary

data. Results: Results revealed small-to-moderate effect sizes, βs = .13 to .22 (p < .05),

indicating that early symptom improvement was related to subsequent alliance quality and that

alliance ratings also were related to subsequent symptom reduction. The relationship between

early alliance and treatment outcome was partially accounted for by early symptom

improvement. With regard to moderators, early alliance showed weaker associations with

outcome in therapies with a strong behavioral component relative to non-behavioral therapies.

However, alliance showed stronger relations to outcome for younger (versus older) patients, over

and above the variance shared with early symptom improvement. Discussion: In sum, early

symptom reduction enhances therapeutic alliance and treatment outcome in EDs, but early

alliance may require specific attention for younger patients and for those receiving non-

behaviorally-oriented treatments.

Keywords: eating disorder, meta-analysis, therapeutic alliance, treatment outcome

of 54




Acc

epte

d A

rtic


5

A Meta-Analysis of the Relation between Therapeutic Alliance and Outcome in Eating Disorders

Therapeutic alliance, defined as the collaborative working relationship between patient

and therapist, is one of the most frequently investigated common factors associated with

psychotherapy outcome (Horvath, Del Re, Flückiger, & Symonds, 2011; Karver, Handelsman,

Fields, & Bickman, 2006; Shirk, Karver, & Brown, 2011). In a meta-analysis of 190 studies of

adult patients with various psychiatric diagnoses, alliance correlated moderately with outcome at

r = .28 (95% confidence interval .25 to .30) (Horvath et al., 2011). A meta-analysis of child and

youth psychotherapy had similar findings, rw1

= .22 (95% confidence interval .16 to .28) (Shirk et

al., 2011). Given the robust association between therapeutic alliance and outcome, researchers

have concluded that alliance is a critical component of effective psychotherapies (Horvath et al.,

2011; Miller & Mizes, 2000; Shirk et al., 2011).

Substantial debate surrounds the importance of therapeutic alliance in eating disorders

(EDs). Although qualitative research has consistently indicated that individuals with EDs find

their relationship with the therapist to be important to their well-being, recovery, and treatment

satisfaction (e.g., Escobar-Koch, Mandlich, & Urzua, 2012), quantitative research on the

relationship between the alliance and outcome in ED treatment has yielded mixed results.

Multiple studies have shown that therapeutic alliance predicts outcome (e.g., Bourion-Bedes et

al., 2013; Constantino, Arnow, Blasey, & Agras, 2005; Zeeck & Hartmann, 2005); yet, other

studies have found little or no association (e.g., Waller, Evans, & Stringer, 2012; Zaitsoff, Doyle,

Hoste, & Le Grange, 2008). Discrepant results across studies may be due to study-level

differences in therapeutic approach, ED diagnosis, patient age, or drop-out.

1 rw = weighted mean correlation

of 54




Acc

epte

d A

rtic


6

The importance of early alliance relative to that of early symptom change in ED

treatment is also unclear. A number of studies have observed strong associations between

symptom change and therapeutic alliance in the first few weeks of treatment (Brown, Mountford,

& Waller, 2013b; Constantino et al., 2005), as well as early symptom change and later outcomes

(Le Grange, Accurso, Lock, Agras, & Bryson, 2014; Raykos, Watson, Fursland, Byrne, &

Nathan, 2013). Thus, it could be argued that the alliance is simply a by-product of early

symptom change, and that alliance-outcome associations that do not account for the role of early

symptom change may be spurious (DeRubeis, Brotman, & Gibbons, 2005). To the extent that a

quality alliance may result from versus promote change, some have questioned whether alliance

is overvalued, and whether its importance may vary by treatment type (Brown, Mountford, &

Waller, 2013a).

Possible Moderators of the Relation between Therapeutic Alliance and Outcome

The strength of the relation between therapeutic alliance and outcome reported in prior

studies may depend on a number of study-level characteristics, including therapy type, mean

patient age, patient diagnosis, alliance rater, and dropout rate.

Therapy type. Findings regarding differences in the relationship between alliance and

treatment outcome for different types of therapy have been inconclusive. In the non-ED

literature, a study investigating two treatments for borderline personality disorder indicated that

alliance was more important for outcome in patients receiving behavioral (i.e., dialectical

behavioral therapy) versus non-behavioral (i.e., community care by experts) treatment (Bedics,

Atkins, Harned, & Linehan, 2015). Conversely, one meta-analysis found that alliance was

relevant to the outcome of therapy only when that therapy was relatively unstructured (i.e., non-

behavioral) (Crits-Christoph et al., 1991); though other meta-analyses have not replicated this

of 54




Acc

epte

d A

rtic


7

distinction (Horvath et al., 2011). In EDs specifically, CBT researchers have questioned the

relationship between alliance and outcome, with certain studies finding no relationship between

alliance and outcome in CBT for anorexia nervosa (AN; e.g., Waller et al., 2012) and bulimia

nervosa (BN; e.g., Raykos et al., 2013).

Patient age. The development of therapeutic alliance may differ in younger versus older

patients. Specifically, child and adolescent patients may have limited abstract reasoning skills

(Bravender et al., 2007), minimize or deny symptoms, or feel pressure from caregivers to enter

treatment involuntarily (Sperry, Roehrig, & Thompson, 2009). Thus some have argued that

clinicians should pay extra attention to establishing a strong alliance relative to other goals early

in youth treatment (Sperry et al., 2009). In line with these suggestions, in studies of child and

adolescent therapy in general, Shirk et al. (2011) found a trend for stronger alliance-outcome

associations among younger patients. In contrast, there is also reason to believe that alliance-

outcome associations might be less important to outcome in youth with EDs. For example,

family-based treatment (FBT), which empowers parents to take charge of their child’s eating,

emphasizes a strong alliance with caregivers early in treatment, which may alter the nature of the

relationship between patient-rated alliance and outcome. In a meta-analysis of youth treatment

studies for a variety of psychiatric disorders, the alliance-outcome association was weaker for

family versus individual therapies (McLeod, 2011). While prior ED studies have separately

focused on patients of different ages, none have examined patient age as a moderator of the

association between alliance and outcome.

Patient diagnosis. Clinicians have posited differences in the overall quality of the

alliance based on ED diagnosis and have speculated that treatment resistance among patients

with AN may hinder the development of a positive alliance (Strober, 2004). However, multiple

of 54




Acc

epte

d A

rtic


8

studies have shown alliance to be relatively strong among patients with AN (Sly, Morgan,

Mountford, & Lacey, 2013; Waller, et al., 2012). In fact, Antoniou and Cooper’s qualitative

review of the relationship between alliance and outcome in EDs (2013) suggested that the

alliance strongly predicted outcome for patients with AN, whereas findings for BN, binge eating

disorder (BED), and subthreshold eating disorders were mixed.

Therapeutic alliance rater. Studies have shown differential effects depending on whether

therapeutic alliance was rated by the patient, the therapist, or an independent observer. In some

studies, patient and independent observer ratings of alliance have shown stronger relationships to

treatment outcome than therapist ratings (Bachelor & Horvath, 1999). In the case of FBT for

EDs, the alliance rating is also complicated by the presence of not only the patient but also the

parents, who are expected to implement important treatment interventions. Differences between

mother-rated, father-rated, and observer-rated alliance and outcome were noted in a study of

FBT for AN (Ellison et al., 2012), with mother-rated alliance showing the strongest relationship

to weight gain. Two different studies analyzing data from a large randomized controlled trial

comparing CBT and IPT for BN (Constantino et al., 2005; Loeb et al., 2005) found that patient-

rated alliance predicted outcome, whereas observer-rated alliance did not.

Drop-out. Drop-out is a substantial problem in ED treatment studies, with attrition rates

ranging from 20-73% in inpatient and outpatient settings (Fassino, Pierò, Tomba, & Abbate-

Daga, 2009). ED research reflects consistent findings from the wider alliance literature,

observing that poor alliance predicts drop-out (Morlino et al., 2007; Sly et al., 2014). Given that

variability in therapeutic alliance is associated with drop-out, it is possible that studies with high

drop-out would show different alliance-outcome associations versus those with low drop-out.

of 54




Acc

epte

d A

rtic


9

Other variables. Other variables, including how therapeutic alliance is measured, and

how treatment outcome is defined, could also impact the relation between alliance and outcome.

The Current Meta-Analysis

Primary Questions. To better understand the relationship between therapeutic alliance

and treatment outcome in EDs, we conducted the first meta-analysis on this topic. Specifically,

we evaluated the aggregated strength of the relationship between alliance and outcome by

conducting temporal analyses of symptom change. Thus change in ED symptoms (i.e., weight,

ED behaviors, and ED cognitions) over the course of treatment was our definition of outcome in

the current meta-analysis. A significant correlation between therapeutic alliance measured at

some point in treatment and a treatment outcome, with no covariates in the model, does not

demonstrate that the alliance is a causal mechanism of symptom change. In this scenario, there is

no control over (1) temporal precedence (i.e., that alliance promotes change measured after

alliance measurement) and (2) the potential role of change occurring prior to alliance

measurement (i.e., the notion that the alliance may be epiphenomenal to symptom reduction that

has already occurred). To better assess whether alliance changes independently from, or in

interaction with, symptom change, we needed to analyze the alliance-outcome association across

multiple points in treatment (with time lags to address temporal sequencing) and account for the

role of prior symptom change (Brown et al., 2013a). Because the data required to perform

temporal analyses were not included in published articles, our team contacted the corresponding

authors of all studies meeting inclusion criteria to acquire the necessary data. Studies whose

author(s) responded to our request and were able to retrieve the needed data were included in our

meta-analysis (see Method).

of 54




Acc

epte

d A

rtic


10

Our analyses addressed four questions. The first three concerned the relationship between

symptom change and later therapeutic alliance at different points in the treatment, and the fourth

addressed the relationship between early alliance and later symptom improvement: (1) Does

early change in symptoms (i.e., early improvement) predict early/mid alliance? (2) Does mid-to-

end of treatment change in symptoms predict alliance at the end of treatment? (3) Does change in

symptoms across the entirety of treatment predict alliance at the end of treatment? (4a) Does

early/mid alliance predict subsequent change in symptoms? And (4b) Do early/mid alliance and

early symptom change each predict unique variance in subsequent change in symptoms (i.e.,

question 4b is an extension of question 4a, but controlling for symptom change)?

Potential moderators. In addition to evaluating the strength of the relationship between

therapeutic alliance and symptom change, we explored potential moderators (i.e., study-level

characteristics that could explain variance in effect sizes). Based on prior literature, we

hypothesized that study-level characteristics including therapy type, patient age, patient

diagnosis, alliance rater, and study drop-out rate would contribute to differences in effect size.

Method

Inclusion Criteria

We set the following inclusion criteria for studies in our meta-analysis: (a) comprised a

sample of patients diagnosed with one or more ED(s), including AN, BN, BED, EDNOS, or sub-

threshold diagnoses; (b) included a measure of therapeutic alliance at one or more time points to

one or more sample groups during the study (e.g., Working Alliance Inventory, Helping Alliance

Questionnaire, Helping Relationship Questionnaire, or California Psychotherapy Alliance

Scales); (c) conducted and reported at least one statistical analysis of the relationship between

alliance and a primary treatment outcome variable (e.g., weight, binge/purge frequency, self-

report or interview measure of ED psychopathology); (d) was not a case report; (e) was

of 54




Acc

epte

d A

rtic


11

published between the dates of January 1978 (i.e., the date of the first ED treatment study to

report alliance-outcome data) and January 2014 (f) was published in English; and (g) did not

utilize data already reported in another study included in the meta-analysis. All studies meeting

each of these requirements were retained for further inspection, while the remaining studies were

assigned reasons for exclusion.

Selection of Studies

To identify relevant studies, we conducted a computer-based search using PsycINFO,

PubMed, and Academic Search Premier. We also searched ProQuest Dissertations and Theses

specifically to locate unpublished studies. We identified search terms for alliance and EDs in the

controlled vocabulary of each database. For example, in PsychINFO, the terms for therapeutic

alliance were alliance, therapeutic alliance, treatment alliance, helping alliance, working

alliance, psychotherapy relationship, therapeutic relationship, therapeutic bond, helping

relationship, and patient therapist relationship. The PsychINFO terms for EDs were eating

disorders, anorexia, bulimia, binge eating disorder, EDNOS, and eating disorder not otherwise

specified. We then searched each database for studies that were tagged with both alliance and

EDs controlled-vocabulary terms. Lastly, we mined the reference section of eight review articles

relevant to alliance in EDs which we identified via the initial electronic database search (Fassino

& Abbate-Daga, 2013; Manlick, Cochran, & Koon, 2013; Martin et al., 2011; Shirk et al., 2011;

Vitousek & Watson, 1998; Vocks et al., 2010; Westwood & Kendal, 2012; Wilson, 2011) for

any additional relevant studies that may have been missed.

The electronic database search combined with the hand search of the review articles

resulted in an initial candidate study pool of 767 studies. These studies were then reduced in a

stepwise fashion by two independent coders (the first and second authors), as described in the

of 54




Acc

epte

d A

rtic


12

PRISMA diagram (Figure 1). The two coders first screened each abstract, applying the a priori

inclusion and exclusion criteria. Of the initial pool of abstracts, 48 studies were retained for full-

text screening. The inter-rater reliability between the two coders for abstract screening was

acceptable with a kappa = .67, p < .01. When the coders’ ratings diverged, they were discussed

until consensus was achieved. The two coders then independently screened the full text of the 48

retained studies. This process resulted in a reduced pool of 27 eligible studies. The inter-rater

reliability between the two coders for the full-text screening was substantial with a kappa = .76, p

<.01. These studies were then back-searched using Google Scholar in order to locate any

additional studies referencing those already included in the pool. None of the new studies located

during this final step met inclusion criteria.

Requests for Additional Data

In order to perform the temporal analyses necessitated by our research questions, our

team contacted the corresponding authors of all 27 eligible studies to request additional—

typically unpublished—data that would be required. We formulated individualized email

requests for each author(s) based on data available from the published report. We received data

from the participating studies between May and October of 2014.

Of the 27 authors who received email requests, 20 responded positively, and were able to

forward all necessary data in a usable format for the proposed temporal analyses. Only six

authors responded negatively to our request, citing that they either (1) did not wish to participate

(Ellison et al., 2012; Hildebrandt, Loeb, Troupe, & Delinsky, 2012; Hoffman, 2006); (2) were

not able to provide the requested data because it was inaccessible (Treasure et al., 1999; Wilson

et al., 1999); or (3) did not collect data from the needed time points (Hartmann, Orlinsky, Weber,

of 54




Acc

epte

d A

rtic


13

Sandholz, & Zeeck, 2010). Finally, one dissertation (Leonard, 2007) could not be included

because we could not locate contact information for the corresponding author.

Measures of Outcome (i.e., ED Symptoms) and Therapeutic Alliance

Outcome (i.e., ED symptoms). In the 20 studies included in our meta-analysis,

investigators measured improvement in ED symptoms with several relevant measures including

body mass index (BMI), weight, percent ideal body weight, binge/purge frequency, vomiting

frequency, body checking frequency, Eating Disorder Examination-Questionnaire (EDE-Q),

Outcome Questionnaire-45.2, and urge to restrict.

Therapeutic alliance. In the 20 studies included in our meta-analysis, investigators

measured therapeutic alliance with nine different scales: Agnew Relationship Measure; Bern

Post-Session Reports for Patients; California Psychotherapy Alliance Scales; Helping Alliance

Questionnaire; Helping Relationship Questionnaire; System for Observing Family Alliances;

Scale for the Multiperspective Assessment of General Change Mechanisms in Psychotherapy;

Treatment Satisfaction Scale; and Working Alliance Inventory. We broadly defined early

alliance as the point in treatment when alliance was first measured. For most studies, this point in

treatment was between sessions 1 and 5 with the exception of one naturalistic longitudinal study

that first measured the alliance at 6 months of treatment, which was approximately mid-way

through therapy (average length of treatment, M = 18 ± 19 months; Paulson Karlsson, Clinton, &

Nevonen, 2013). We defined mid alliance as the point at which the alliance rating occurring

closest to the midpoint of treatment. For most studies, mid alliance was measured between

session 6 and 12. We defined late alliance as the alliance rating at the end of treatment. This was

always the last alliance measurement taken; timing varied across studies.

Levels of Each Moderator Variable

of 54




Acc

epte

d A

rtic


14

We examined the following variables as possible moderators of the alliance-outcome

effect size, classifying each study as falling into one of the following levels on each moderator.

Therapy type. We coded therapy type as a categorical variable with five categories:

behavioral weight-loss therapy (BWLT), CBT, FBT, individual-focused therapy, or multiple

therapies. The BWLT category comprised a manualized behavioral treatment following the

tetrahydrolipstatin-based weight loss manual that focuses on balanced nutrition and physical

activity to promote weight loss (Margraf, 2000; Munsch, Biedert, & Keller, 2003). The CBT

category included manualized treatments that employ both cognitive and behavioral strategies to

promote eating-disorder symptom change. The FBT category included a manualized treatment

that empowers parents to effect change in their child’s eating-disorder symptoms (Lock, Le

Grange, Agras, & Dare, 2001). The individual-focused therapy category included therapies that

fostered the development of insight in related areas, but did directly encourage change in eating-

disorder symptoms, including adolescent-focused therapy (AFT), IPT, and supportive

psychotherapy (SPT). The multiple therapies category included studies in which participants

received two or more different types/modes of therapy either simultaneously or consecutively

(e.g., a mixture of inpatient, day-patient, outpatient, individual, group, and/or family therapies

[Paulson Karlsson et al., 2013], a treatment combining individual therapy and a supportive

program aimed at improving weight and eating behaviors [Bourion-Bedes et al., 2013]).

Mean age. We recorded the mean age of each study sample as a continuous variable.

When there was more than one sample in a study (e.g., a randomized controlled trial), we

recorded the mean age for each sub-sample. However, when the mean age for each subsample

was not reported and there was no statistically significant difference in mean age between the

sub-samples, the mean age for the total sample was used for both subsamples.

of 54




Acc

epte

d A

rtic


15

ED diagnosis. We coded ED diagnosis as a categorical variable with four categories:

AN, BN, BED, or multiple. A sample was coded as multiple when the sample was composed of

people with different ED diagnoses.

Therapeutic alliance rater. We coded therapeutic alliance rater as a categorical variable

with two categories: patient-rated or independent observer-rated. In instances where data from

more than one rater of the alliance was included in a study (e.g., patient-rated and parent-rated

alliance or patient-rated and therapist-rated alliance), we chose to use the patient-rated alliance or

the independent observer-rated alliance (if patient-ratings were not provided) because patient

ratings and independent observer ratings have shown stronger associations to treatment outcome

than parent or therapist ratings of the alliance (Bachelor & Horvath, 1999; Horvath et al., 2011).

We did not have any studies in our pool that only collected therapist- or parent-rated alliance

data.

Study drop-out rate. We recorded study drop-out rate as a continuous variable. When

there was more than one sample in a study (e.g., a randomized controlled trial), each sub-sample

was assigned the same study drop-out rate, unless drop-out was reported individually for each

sub-sample.

Effect Size Information

We used the standardized regression coefficient (β) to evaluate effect size for each of our

four meta-analytic research questions. Rather than extracting effect-size data from the original

papers, we obtained more detailed information (i.e., descriptive statistics and correlations)

directly from the authors of each study—this was necessary because many studies did not report

the information needed to calculate temporal effect sizes. When there was missing data in the

summary statistics, we used pairwise deletion in the analyses required to obtain the effect sizes

of interest to increase sample size and thus power. When there was no missing data reported in

of 54




Acc

epte

d A

rtic


16

the summary statistics, analyses required to obtain the effect sizes of interest were based on the

total sample.

To facilitate comparison across therapy types within each study, we calculated separate

effect sizes of the alliance-outcome relation for each treatment arm. We then calculated our own

effect sizes, standard errors (SE), 95% confidence intervals (CI), and p-values using individual

multiple linear regression analyses in SPSS. In all analyses, we coded treatment outcome

variables such that higher positive scores indicated greater symptom change (e.g., increased BMI

in AN trials, decreased binge/purge frequency in BN trials) and stronger alliance. Some studies

utilized multiple measures of the same construct—either alliance or ED symptoms. When a

particular measure included more than one subscale that could be combined into a global score

(e.g., EDE-Q), we used the global score to calculate the effect size for the relation between

alliance and outcome. When a measure contained subscales that could not be combined to

achieve a total score (e.g., Working Alliance Inventory), we averaged the effect sizes for the

outcome-alliance relation for each subscale, to obtain an average effect size for that study (as in

Thomas, Vartanian, & Brownell, 2009).

For the first question (i.e., Does early symptom change predict early/middle alliance?),

the regression analysis included (1) ED symptoms at baseline, and (2) change in symptoms from

baseline to when alliance was first measured, as predictors of the first measure of alliance. Thus,

the standardized regression coefficient indexed the relationship between early symptom change

and alliance, controlling for baseline symptom level. For the second question (i.e., Does middle

to end of treatment symptom change predict later alliance?), the regression analysis included (1)

ED symptoms when alliance was first measured and (2) change in symptoms from when alliance

was first measured to the end of treatment as predictors of alliance at the end of treatment.

of 54




Acc

epte

d A

rtic


17

Therefore, the standardized regression coefficient indexed the relationship between change in

treatment outcome and alliance at the end of treatment, controlling for symptom level at the time

of first alliance measurement. For the third question (i.e., Does symptom change across treatment

predict later alliance?), the regression analysis included (1) ED symptoms at baseline and (2)

change in symptoms from baseline to end of treatment as predictors of alliance at the end of

treatment. Thus, the standardized regression coefficient indexed the relationship between change

in treatment outcome from baseline to end of treatment and alliance at the end of treatment,

controlling for baseline symptom level. For the fourth question (i.e., Does early/mid alliance

predict subsequent symptom change?), the regression analysis included (1) alliance and (2) ED

symptoms when the alliance was first measured as predictors of change in treatment outcome

from when the alliance was first measured to the end of treatment. Therefore, the standardized

regression coefficient indexed the relationship between early/mid alliance and subsequent

symptom change, controlling for symptom level at the time of alliance measurement. We also

conducted a second regression analysis to examine whether early/mid alliance predicts

subsequent symptom change above and beyond early symptom change. Thus, the regression

analysis included (1) the first measure of alliance and (2) change in symptoms from baseline to

when alliance was first measured as predictors of change in treatment outcome from when the

alliance was first measured to the end of treatment. The standardized regression coefficient

indexed the relationship between early/mid alliance and subsequent symptom change while

statistically controlling for early symptom change.

Meta-analytic Procedures

of 54




Acc

epte

d A

rtic


18

For each research question, we pooled relevant effect sizes, weighted by their inverse-

variance (1/SE2). The SE of each effect size (β) was calculated using the formula provided by

Cohen, Cohen, West, & Aiken (2003; also see Card, 2013):

�� = � 1–�� − � − 1� 11 −�� where ��is the variance explained in the dependent variable by the independent variables in the

regression model, ��is the variance explained in the independent variable of interest by the

remaining independent variables in the regression model, n is the sample size, and k is the

number of independent variables in the regression model. We interpreted the magnitude of each

effect size according to Cohen’s (1988) conventions for correlation coefficients, where .10 is

small, .30 is moderate, and .50 is large.

To allow us to generalize our results beyond the current sample, we used a random-

effects model. We assessed publication bias using Egger’s test (Egger, Davey, Schneider, &

Minder, 1997), which examines the presence of asymmetry in a funnel plot of effect sizes. To

examine the impact of each individual effect size on the overall mean effect size, we also

conducted a one study removed sensitivity analysis for each meta-analytic research question.

Furthermore, we assessed whether the effect sizes were more heterogeneous than expected by

sampling variability alone using the test of heterogeneity (Q-statistic). When there was evidence

of heterogeneity, we used the I2 statistic to quantify the extent of heterogeneity. We then

conducted follow-up moderator analyses using random-effects analogue to ANOVA for

categorical moderators, and random-effects meta-regression for continuous moderators. When

one or more statistically significant moderators were at least moderately correlated, we

conducted a meta-regression analysis in which we controlled for their shared association. We

of 54




Acc

epte

d A

rtic


19

conducted all analyses using Comprehensive Meta-Analysis Version 2.0 software program

(Borenstein, Hedges, Higgins, & Rothstein, 2005) except for the meta-regressions which we

conducted using SPSS macros (Lipsey & Wilson, 2001).

Results

(1) Does early symptom change predict early/mid alliance?

Omnibus test. A total of 18 independent effect sizes from 14 different reports evaluated

the relationship between early symptom change and early/mid therapeutic alliance (Table 1). For

most reports, early/mid alliance was measured between sessions 3 and 10 of treatment; one

naturalistic longitudinal study first measured the alliance after 6 months of treatment, which was

approximately mid-way through therapy in this particular design (average length of treatment, M

= 18 ± 19 months) (Paulson Karlsson et al., 2013). As expected, greater positive change in

symptoms (i.e., greater improvement) from baseline to when the alliance was first measured

predicted stronger early/mid alliance, β = .19, 95% CI [.11, .28], z = 4.38, p < .0001. The

magnitude of the mean effect size was small-to-moderate and there was no evidence of

publication bias, Egger’s regression intercept = .02, t (16) = .06, p = .95. The mean effect size

was stable in our one study removed sensitivity analysis, ranging from .17 to .22.

Moderator analyses. In addition, the effect sizes were heterogeneous, Q (17) = 28.41 p =

.04, but the extent of heterogeneity was low, I2

= 40.16. In follow-up moderator analyses, study

drop-out rate was associated with effect sizes at trend-level, Q (1) 3.65, p = .06. Specifically,

studies with higher drop-out rates had larger effect sizes, slope = .01, 95% CI [-.0002, .014], z =

1.91, p = .06. To further evaluate this finding, we examined the mean effect size at high (+1 SD)

and low (-1 SD) levels of study drop-out rate (weighted M = 14.60% drop-out rate, SD = 5.19).

At 1 SD above the mean of study drop-out rate, the effect size was small-to-moderate and

of 54




Acc

epte

d A

rtic


20

statistically significant, β = .21, 95% CI [.13, .30], z = 4.91, p < .001. Likewise, at 1 SD below

the mean of study drop-out rate, the effect size was small and statistically significant, β = .14,

95% CI [.04, .24], z = 2.84, p = .004. Taken together, these findings indicate a positive linear

relationship between the magnitude of the effect sizes and study drop-out rate. None of the

remaining moderators were statistically significant (Table 2).

(2) Does mid-to-end of treatment change in symptoms predict later alliance?

Omnibus test. A total of ten independent effect sizes from eight different reports

evaluated the relationship between mid-to-end of treatment symptom change and later

therapeutic alliance (see Table 3). Alliance in all reports was measured at the end of treatment

(i.e., at the final treatment session). Results for the overall mean effect size indicated that change

in symptoms (i.e., improvement) from when early/mid alliance was measured until the end of

treatment was not related to later alliance, β = .10, 95% CI [-.04, .24], z = 1.46, p = .15. The

mean effect size was not statistically significant and there was no evidence of publication bias,

Egger’s regression intercept = .02, t (8) = .02, p = .97. The mean effect size was stable in our one

study removed sensitivity analysis, ranging from .07 to .12.

Moderator analyses. Because the effect sizes were homogenous, Q (9) = 2.79, p = .97,

we did not evaluate potential moderators.2

(3) Does change in symptoms across treatment predict later alliance?


the relationship between change in symptoms across treatment and later alliance (Table 4). In

almost all reports, later alliance was measured at the end of treatment. As expected, greater

2 Although moderator analyses are often underpowered in meta-analyses comprising a relatively small

number of studies, we chose to remain conservative by following the recommendations of Cooper,

Hedges, & Valentine (2009) and Lipsey and Wilson (2001) not to evaluate moderators following a non-

significant omnibus test.

of 54




Acc

epte

d A

rtic


21

positive change in symptoms (i.e., improvement) across each study’s duration predicted greater

subsequent alliance, β = .17, 95% CI [.06, .29], z = 2.96, p = .003. The mean effect size was

small-to-moderate and there was no evidence of publication bias, Egger’s regression intercept =

.81, t (16) = .94, p = .36. The mean effect size was stable in our one study removed sensitivity

analysis, ranging from .16 to .20.

Moderator analyses. Effect sizes were homogenous, Q (17) = 24.17, p = .12, so we did

not evaluate moderators.

(4a) Does early/mid alliance predict subsequent symptom change?


the relationship between early/mid therapeutic alliance and change in symptoms from when

early/mid alliance was measured to the last time-point of data on symptoms available in each

report (see Table 5). For almost all reports, the last time-point of data on symptoms was the end

of treatment; in one naturalistic longitudinal study, alliance was assessed after 6 months of

treatment3 (Paulson Karlsson et al., 2013). As expected, greater early/mid alliance predicted

greater subsequent symptom change, β = .13, 95% CI [.05, .22], z = 3.10, p = .002. The

magnitude of the mean effect size was small and there was no evidence of publication bias,

Egger’s regression intercept = .48, t (17) = .67, p = .51. The mean effect size was stable in our

one study removed sensitivity analysis, ranging from .11 to .15.

Moderator analyses. The effect sizes were heterogeneous, Q (18) = 26.55, p = .09, and

the extent of heterogeneity was low, I2

= 32.20. As such, we evaluated potential moderators.

Therapy type was related to effect size, Q (4) = 10.61, p = .03. Specifically, greater early/mid

3 While Paulson Karlsson et al. (2013) also measured alliance 36 months after the start of treatment, we

did not include those data in the current meta-analysis because the length of follow-up at the final time

point of this longitudinal study differed so greatly from the other included studies, which were primarily

much briefer randomized controlled trials.

of 54




Acc

epte

d A

rtic


22

alliance predicted greater subsequent positive change in treatment outcome for studies involving

multiple therapies (β = .18, 95% CI [.05 .32], z = 2.70, p = .007, k = 6); individual-focused

therapies (β = .21, 95% CI [.05, .37], z = 2.56, p = .01 k = 4); and FBT (β = .31, 95% CI [.08,

.54], z = 2.62, p = .009, k = 3). In contrast, early/mid alliance was not related to subsequent

positive change in treatment outcome for studies involving BWLT (β = -.05, 95% CI [-.20, .11],

z = -.59, p = .56, k = 2), and CBT (β = -.02, 95% CI [-.25, .21], z = -.19, p = .85, k = 4). A

follow-up meta-regression analysis evaluated mean differences in effect sizes as a function of

therapy type. In the meta-regression, we used CBT as the reference group for therapy type

(BWLT was not suitable to serve as the reference group because there were only two studies).

Therapy type accounted for 43% of the variance in the effect sizes, R2 = .43, Q (4) = 10.35, p =

.04. The mean effect size for studies involving FBT were larger than the mean effect size for

studies involving CBT, B = .31, z = 2.12, p = .03. Likewise, there was a non-statistically

significant trend indicating that the mean effect for studies involving multiple therapies tended to

be larger than the mean effect size for studies involving CBT, B = .18, z = 1.78, p = .07.

Similarly, there was also a non-statistically significant trend indicating that the mean effect size

for studies involving individual-focused therapies tended to be larger than the mean effect size

for studies involving CBT, B = .21, z = 1.76, p = .07. Also, the mean effect size for studies

involving BWLT did not differ from the mean effect size for studies involving CBT, B = -.04, z

= -.37, p = .71, although in this case the size and direction of the effect did not reflect a similar

pattern to the other variables, i.e., it was more similar to the results for CBT. In sum, greater

early/mid alliance predicted greater subsequent positive change in treatment outcome for studies

involving FBT, multiple therapies, and individual-focused therapies relative to studies involving

of 54




Acc

epte

d A

rtic


23

CBT where there was no such effect. None of the remaining moderators were related to

variability in the effect sizes (Table 2).

(4b) Does early/mid alliance predict subsequent symptom change above and beyond early

symptom change?

Omnibus test. A total of 15 independent effect sizes from 11 different reports allowed us

to evaluate the relationship between early/mid therapeutic alliance and change in symptoms from

when early/mid alliance was measured to the last time-point of data on symptoms available in

each report while statistically controlling for early symptom change (Table 6). For almost all

reports, the last time-point of data on symptoms was at end of treatment; however, for one study,

the last time-point of data on symptoms was at 6-month follow-up (Paulson Karlsson et al.,

2013). The mean effect size was not statistically significant, β = .07, 95% CI [-.04, .17], z = 1.26,

p = .21, and there was no evidence of publication bias, Egger’s regression intercept = .09, t (13)

= .09, p = .93. The mean effect size was stable in our one study removed sensitivity analysis,

ranging from .03 to .09.

Moderator analyses. The effect sizes were heterogeneous, Q (14) = 23.15, p = .058, and

the extent of heterogeneity was low, I2 = 39.52%. Thus, we evaluated potential moderators.

Sample mean age was related to effect size, Q (1) = 16.20, p < .01. Specifically, studies with

older samples had smaller effect sizes relative to studies with younger samples, B = 03, z = -4.03,

p < .001. To further evaluate this finding, we examined the mean effect size at high (+1 SD) and

low (-1 SD) levels of sample mean age (weighted M = 22.08 years old, SD = 5.94). At 1 SD

above sample mean age, the effect size was small and was not statistically significant, β = -.10,

95% CI [-.20, .01], z = -1.71, p = .09. This finding indicates that early/mid alliance did not

predict change in symptoms from when the early/mid alliance was measured to the end of

of 54




Acc

epte

d A

rtic


24

treatment above and beyond early symptom change in studies with older patients. However, at 1

SD below the sample mean age, early/mid alliance predicted greater improvement in symptoms

from when the early/mid alliance was measured to the end of treatment above and beyond early

symptom change in studies with younger patients, β = .22, 95% CI [.11, .33], z = 3.98, p = .0001.

The magnitude of the effect size was small-to-moderate.

ED diagnosis was also related to the effect sizes, Q (2) = 6.10, p = .04. Specifically, the

mean effect size was statistically significant and small for studies with samples of AN (β = .16,

95% CI [.04, .27], z = 2.57, p = .01), but was not significant for studies with samples of BN (β =

-.10, 95% CI [-.26, .06], z = -1.17, p = .24) and studies with mixed ED samples (β = .06, 95% CI

[-.18, .30], z = .49, p = .62). In short, early/mid alliance predicted greater improvement in

symptoms from when the early/mid alliance was measured to the end of treatment above and

beyond early symptom change in studies with samples of AN. None of the remaining moderators

were statistically significant (Table 2).

Sample mean age (R2 = .53, p = .04) and ED diagnosis (Cramer’s V = .52, p < .001) were

both moderately associated with therapy type. However, sample mean age and ED diagnosis

were not related (R2 = .14, p = .33). Thus, we conducted a follow-up meta-regression to examine

whether sample mean age and ED diagnosis remained statistically significant predictors of effect

size while controlling for shared variance with therapy type. In the meta-regression, we used AN

as the reference group for ED diagnosis and individual-focused therapies as the reference group

for therapy type. Mean age, ED diagnosis, and therapy type together accounted for 88% of the

variance in the effect sizes, R2 = .88, Q (6) = 20.42, p = .002. Sample mean age accounted for

unique variance in the effect sizes above and beyond ED diagnosis and therapy type, B = -.03, z

= -3. 01, p = .003. In contrast, differences in ED diagnosis did not account for unique variance in

of 54




Acc

epte

d A

rtic


25

the effect sizes above and beyond mean age and therapy type. Differences in therapy type did not

account for variance in the effect sizes. In sum, findings indicated that early/mid alliance

predicted greater subsequent improvement in symptoms above and beyond early symptom

change in studies with younger patients, regardless of their ED diagnosis and therapy type.

Discussion

Although ED clinicians have long stressed the role of therapeutic alliance in facilitating

symptom change, ED researchers studying behavioral treatments have instead stressed the

importance of early symptom change for promoting therapeutic alliance and have debated the

relative and temporal influences of these two factors on each other and outcome. Our meta-

analysis of 20 ED treatment studies, examining the relations between symptom change and

alliance across time and samples, supports a reciprocal relationship between symptom change

and alliance. In addition, our analyses are unique in that they are the first in ED treatments to

identify that the relative importance of therapeutic alliance for treatment outcome may differ

across treatment type, patient age, and patient diagnosis. Interestingly, alliance rater

(independent rater versus patient) did not impact effect sizes. Further, the current study

succeeded in connecting multiple well-known research groups in the field of EDs from across the

globe, representing data from nine different countries. We evaluated four distinct research

questions, finding statistically significant results for three of the four, with all effect sizes being

in the hypothesized direction.

We identified the strongest association between symptom change and subsequent

alliance, specifically a small-to-moderate sized relationship between early symptom change and

early/mid alliance (question 1), as well as a small-to-moderate relationship across-treatment

symptom change and subsequent alliance (question 3). This relationship between symptom

change and alliance early in therapy was not moderated by treatment type, ED diagnosis, or other

of 54




Acc

epte

d A

rtic


26

factors, and therefore should be assumed to hold across all levels of these moderator variables.

The finding that positive symptom change strengthens therapeutic alliance is consistent with

evidence from other psychological disorders, including depression (Tang & DeRubeis, 1999).

However, additional analyses also supported the temporal role of the early alliance in

facilitating later symptom change. Results for question 4a (Does early/mid alliance predict

subsequent symptom change?) indicated that early/mid alliance ratings also predicted subsequent

changes in outcome. Although differences were noted in the relationship between early alliance

and later symptom change between different types of treatment, these results were only

significant at the trend level, and should therefore be interpreted with caution. The results of

moderator analyses supported the role of early alliance in predicting later symptom-change for

individual-focused therapies (e.g., IPT, AFT, and SPT), FBT, and multiple therapies; but not for

CBT or BWLT. Further, meta-regression to explore individual comparisons indicated that the

differences between CBT and other treatments, excepting BWLT, were particularly strong.

These results are very interesting in light of the importance of early symptom change to outcome

in CBT (Brown et al., 2013b). It is possible that the alliance is particularly critical in therapies

where it is viewed and cultivated as an agent of change; however, further research is needed to

confirm this.

Unique analyses compared the relative strength of the associations between treatment

outcome and (a) early therapeutic alliance, and (b) early symptom change, including moderator

analyses to explore potential differences according to patient age and patient diagnosis. The

results indicated that the early alliance was significantly related to subsequent symptom change

for younger patients and for patients with AN, but not for older patients or those with other ED

diagnoses. Further analyses controlling for the correlations between patient age, ED diagnosis,

of 54




Acc

epte

d A

rtic


27

and treatment type, found that patient age produced the only statistically significant effect after

controlling for ED diagnosis and therapy type, indicating that it was a particularly important

predictor of a stronger association between the early alliance and outcome. These findings reflect

the observations of some alliance researchers outside the field of EDs (Shirk et al., 2011), as well

as clinicians who treat adolescents. Importantly, age was a significant moderator even after

controlling for treatment type (i.e., individual versus family-based), suggesting that extra

attention may need to be paid to the alliance relative to other goals early in treatment for younger

patients with EDs (Sperry et al., 2009), regardless of theoretical orientation. Of course, given that

age was examined as a study-level (rather than individual-level) moderator in this meta-analysis,

we can only draw conclusions about studies that recruited younger patients, rather than any

specific youth patient, or youth patients in general.

Our findings also suggested that drop-out rate should be considered when interpreting the

size and significance of the relation between symptom change and early alliance ratings. Results

indicated that when drop-out was low, symptom change showed a smaller relationship to

early/mid alliance ratings, whereas when it was high, early improvement more strongly predicted

early/mid alliance ratings. Patients drop out of studies for a wide range of reasons and at various

points in treatment (both early and late). Studies that retain patients who are otherwise likely to

drop out may include patients with a variety of factors influencing both alliance and symptom

change, introducing other sources of variance and error into the symptom change/alliance

relationship. It is also possible that patients who drop out of treatment tend to have lower levels

of the alliance at the outset or are initially less symptomatic. Thus, it could be argued that drop-

out, outcome, and therapeutic alliance are confounded. This possibility should temper the

of 54




Acc

epte

d A

rtic


28

interpretation that early symptom change predicts later alliance as well. A more nuanced study of

the drop-out/alliance/outcome relationships in ED samples would help to clarify these questions.

Within the 20 studies included in this meta-analysis, nine different measures of

therapeutic alliance were used. Due to the diverse range of alliance measures, it was not possible

to include this variable in our moderator analyses. Research indicates that the shared variance

among the numerous measures of therapeutic alliance is less than 50%, even among the four so-

called “core” measures (i.e., Working Alliance Inventory, Helping Alliance Questionnaire,

California Psychotherapy Alliance Scale, and Vanderbilt Psychotherapy Process Scale)

(Horvath, 2009). This suggests that these scales may all be measuring slightly different

constructs. Future research should be designed to investigate if and how the type of alliance

rating measure used may affect the resulting alliance-outcome associations.

This study has limitations that should be noted. First, our sample of included studies was

relatively small. Although there has been an increased focus on therapeutic alliance in recent

years, there are still relatively few treatment studies within the ED field that have collected both

alliance and outcome data. Moreover, of the studies that have collected such data, most only

assess these variables a few times across treatment. In order to truly begin to untangle this issue,

alliance and outcomes should be measured repeatedly, from session 1 to end of treatment. Our

findings, combined with others from the ED field (i.e., Tasca & Lampard, 2012), suggest that

alliance and outcomes are not static constructs. They change over time and it is quite possible

that it is the change in these constructs that is key. Moreover, although our meta-analysis

provided the unique opportunity to evaluate changes in both alliance and symptoms over time,

the temporal precedence of one over the other does not necessarily imply causality.

of 54




Acc

epte

d A

rtic


29

Further, despite our best attempts to locate all relevant studies and contact all

corresponding authors, there remained a number of applicable studies that were excluded from

our meta-analyses because of (1) difficulties contacting the corresponding author(s), or (2) the

inability of corresponding author(s) to retrieve the needed data. The inclusion of these missing

data could have yielded different results. Second, with regard to our moderator analyses, it is

important to note that study sample sizes (k) for many of these analyses were quite small, and

therefore, results from these analyses should be interpreted with caution. This is particularly true

of the moderator analyses involving therapy type. Third, the majority of the included studies

were composed of Caucasian females (~90%), which greatly reduces the generalizability of our

results to only one subset of the population receiving ED treatment. It is not yet known whether

these results would apply to males and/or patients from ethnically diverse backgrounds. In fact,

one meta-analysis investigating the moderating effects of the presence of racial/ethnic minorities

on the strength of the alliance-outcome association, found that the percentage of overall

minorities (particularly African Americans) attenuated the alliance-outcome association

(Flückiger et al., 2013). Unfortunately, due to largely homogenous study samples in terms of

race and gender and a lack of data regarding patient comorbidities (e.g., substance use disorders),

we were unable to investigate the moderating impact of these variables. Fourth, other patient

variables (e.g., personality characteristics, attachment style) and therapist characteristics (e.g.,

gender, experience level) that may impact both alliance and outcome were not measured in a

sufficient number of studies to be included as potential moderators. A final limitation of the

current meta-analysis is that it was impossible to exclude all third-variable confounds. For

instance, it is plausible that patient characteristics not accounted for in our analyses, such as high

interpersonal functioning, patient level of insight, or patient motivation or expectancies for

of 54




Acc

epte

d A

rtic


30

change, are associated with both greater alliance and outcome (Jones, Lindekilde, Lübeck, &

Clausen, 2015).

Conclusions

Overall, the bidirectional relationship between therapeutic alliance and outcome found in

our meta-analysis strongly suggests the critical value of both early and sustained symptom

change, as well as the patient-therapist relationship in this clinically challenging population.

Symptom improvement was shown to predict subsequent alliance both early in and across the

span of treatment, irrespective of treatment type, patient age, or ED diagnosis. Differences in the

strength of the relationship between the early alliance and treatment outcome were observed for

different treatments, with CBT and BWLT showing weaker associations than other treatment

types. Multivariate analyses examining the relative strength of associations between early

alliance and later outcome controlling for early symptom change, and examining differences in

these relationships according to patient age and patient diagnosis, found that early symptom

change accounts for a moderate portion of observed associations between the early alliance and

outcome. Analyses indicated that for older patients and those with BN, BED, and

mixed/subclinical diagnoses, attention to early symptom change may yield the most benefit for

both the early alliance and eventual treatment outcome. However, results of these analyses

indicated that younger patients may show specific benefit from additional attention to the early

alliance, which showed associations with outcome even when early symptom change was taken

into account. These results support a more fine-grained and complex approach to research

concerning the inter-relationships between symptom improvement, alliance, and treatment

outcome, with attention paid to possible differences in these relationships according to treatment

approach and patient factors. Further research is needed to determine the extent to which the

of 54




Acc

epte

d A

rtic


31

bidirectional relationship between therapeutic alliance and symptom change and its attendant

moderators is unique to EDs, or more broadly applicable across psychiatric disorders.

of 54




Acc

epte

d A

rtic


32

References

References marked with an asterisk indicate studies included in the meta-analysis.

Antoniou, P. & Cooper, M. (2013). Psychological treatments for eating disorders: What is the

importance of the quality of the therapeutic alliance for outcomes? Counselling

Psychology Review, 28, 34-46.

Bachelor, A. & Horvath, A. (1999). The therapeutic relationship. In M. A. Hubble, B. L.

Duncan, & S. D. Miller (Eds).The heart and soul of change: What works in therapy (pp.

133-178). Washington, DC: American Psychological Association.

Bedics, J. D., Atkins, D. C., Harned, M. S., & Linehan, M. M. (2015). The therapeutic

alliance as a predictor of outcome in dialectical behavior therapy versus non-behavioral

psychotherapy by experts for borderline personality disorder. Psychotherapy, 52, 67-

77. doi.org/10.1037/a0038457

Bordin, E. S. (1979). The generalizability of the psycho-analytic concept of the working alliance.

Psychotherapy: Theory, Research, and Practice, 16, 252–260.

Borenstein, M., Hedges, L., Higgins, J., & Rothstein, H. (2005). Comprehensive meta-analysis:

Version 2.0. Englewood, NJ. Biostat.

*Bourion-Bedes, S., Baumann, C., Kermarrec, S., Ligier, F., Feillet, F., Bonnemains, C.,

Guillemin, F., & Kabuth, B. (2013). Prognostic value of early therapeutic alliance in

weight recovery: A prospective cohort of 108 adolescents with anorexia nervosa. Journal

of Adolescent Health, 52, 344–350. doi.org/10.1016/j.jadohealth.2012.06.017

Bravender, T., Bryant-Waugh, R., Herzog, D., Katzman, D., Kreipe, R. D., Lask, B., ... &

Nicholls, D. (2007). Classification of child and adolescent eating disturbances.

of 54




Acc

epte

d A

rtic


33

Workgroup for Classification of Eating Disorders in Children and Adolescents

(WCEDCA). The International Journal of Eating Disorders, 40, S117-22.

Brown, A., Mountford, V. A., & Waller, G. (2013a). Is the therapeutic alliance overvalued in the

treatment of eating disorders? International Journal of Eating Disorders, 46, 779-782.

doi: 10.1002/eat.22177

*Brown, A., Mountford, V. A., & Waller, G. (2013b). Therapeutic alliance and weight gain

during cognitive behavioural therapy for anorexia nervosa. Behavior Research and

Therapy, 51, 216–220. doi.org/10.1016/j.brat.2013.01.008

Card, N. A. (2013). Applied meta-analysis for social science research. New York, NY: Guilford

Press.

Cohen, J. (1988). Statistical Power Analysis for the Behavioral Sciences. New York, NY:

Routledge Academic.

Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2003). Applied Multiple

Regression/Correlation Analysis for the Behavioral Sciences (3rd

ed.). Mahwah, New

Jersey: Lawrence Erlbaum Associates, Publishers.

*Constantino, M. J., Arnow, B. A., Blasey, C., & Agras, S. W. (2005). The association between

patient characteristics and the therapeutic alliance in cognitive-behavioral and

interpersonal therapy for bulimia nervosa. Journal of Consulting and Clinical

Psychology, 73, 203–211. doi: 10.1037/0022-006X.73.2.203

Crits-Christoph, P., Baranackie, K., Kurcias, J. S., Beck, A.T., Carroll, K., Perry, K.,… Zitrin, C.

(1991). Meta-analysis of therapist effects in psychotherapy outcome studies.

Psychotherapy Research, 1, 81–91. doi.org/10.1080/10503309112331335511

DeRubeis, R. J., Brotman, M. A., & Gibbons, C. J. (2005). A conceptual and methodological

of 54




Acc

epte

d A

rtic


34

analysis of the nonspecific argument. Clinical Psychology: Science and Practice, 12,

174-183. doi:10.1093/clipsy/bpi022

Egger, M., Davey, S., Schneider, M., & Minder, C. (1997). Bias in meta-analysis detected by a

simple, graphical test. British Medical Journal, 315, 629-634.

doi.org.ezproxy.lib.umb.edu/10.1136/bmj.315.7109.629

Ellison, R., Rhodes, P., Madden, S., Miskovic, J., Wallis, A., Baillie, A.,… Touyz, S. (2012). Do

the components of family based treatment for anorexia nervosa predict weight gain?

International Journal of Eating Disorders, 45, 609-614. doi: 10.1002/eat.22000

Escobar-Koch, T., Mandlich, C. C., & Urzua, R. F. (2012). Treatments for eating disorders: The

patients’ views. In I.J. Lobera (Ed.), Relevant topics in eating disorders, Prof. Ignacio

Jáuregui Lobera (Ed.), ISBN: 978-953-51-0001-0, InTech, Available from:

http://www.intechopen.com/books/relevant-topics-in-eatingdisorders/ treatments-for-

eating-disorders-the-patients-views

Fassino, S. & Abbate-Daga, G. (2013). Resistance to treatment in eating disorders: A critical

challenge. BMC Psychiatry, 13, 1-18. doi: 10.1186/1471-244X-13-282

Fassino, S., Pierò, A., Tomba, E., & Abbate-Daga, G. (2009). Factors associated with drop-out

from treatment for eating disorders: A comprehensive literature review. BMC Psychiatry,

9, 1-9. doi:10.1186/1471-244X-9-67

Flückiger, C., Horvath, A. O., Ackert, M., Del Re, A. C., Symonds, D., & Wampold, B. E.

(2013). Substance use disorders and racial/ethnic minorities matter: A meta-analytic

examination of the relation between alliance and outcome. Journal of Counseling

Psychology, Advance online publication. doi: 10.1037/a0033161.

of 54




Acc

epte

d A

rtic


35

*Flückiger, C., Meyer, A., Wampold, B. E., Gassmann, D., Messerli-Bürgy, N., & Munsch, S.

(2011). Predicting premature termination within a randomized controlled trial for binge-

eating patients. Behavior Therapy, 42, 716–725.

*Forsberg, S., LoTempio, E., Bryson, S., Fitzpatrick, K. K., Le Grange, D., & Lock, J. (2013).

Therapeutic alliance in two treatments for adolescent anorexia nervosa. International

Journal of Eating Disorders, 46, 34–38. doi: 10.1002/eat.22047

Hartmann, A., Orlinsky, D., Weber, S., Sandholz, A. & Zeeck, A. (2010). Session and

intersession experience related to treatment outcome in bulimia nervosa. Psychotherapy,

Theory, Research, Practice, Training, 47(3), 355-370. doi: 10.1037/a0021166.

Hildebrandt, T., Loeb, K., Troupe, S., & Delinsky, S. (2012). Adjunctive mirror exposure for

eating disorders: A randomized controlled pilot study. Behaviour Research and Therapy,

50, 797-804. doi.org/10.1016/j.brat.2012.09.004

Hoffman, J. E. (2006). Alliance as a predictor of outcome in group psychotherapy for binge

eating disorder: Making the most predictive model through exploration of mediation and

moderation (Doctoral dissertation). Retrieved from Dissertation Abstracts International:

Section B: The Sciences and Engineering. (2009-99040-372)

Horvath, A. O. (2009, October). Conceptual and methodological challenges in alliance research

Is it time for a change? Paper presented at the Society for Psychotherapy Research

European Division, Bolzano, Italy.

Horvath, A. O., Del Re, A. C., Flückiger, C., & Symonds, D., (2011). Alliance in individual

psychotherapy. Psychotherapy, 48, 9-16. doi: 10.1037/a0022186

*Isserlin, L. & Couturier, J. (2012). Therapeutic alliance and family-based treatment for

adolescents with anorexia nervosa. Psychotherapy, 49, 46–51. doi: 10.1037/a0023905

of 54




Acc

epte

d A

rtic


36

Jones, A., Lindekilde, N., Lübeck, M., & Clausen, L. (2015). The association between

interpersonal problems and treatment outcome in the eating disorders: A systematic

review. Nordic Journal of Psychiatry,13, 1-11. [Epub ahead of print]

Karver, M., Handelsman, J., Fields, S., & Bickman, L. (2006). Meta-analysis of relationship

variables in youth and family therapy: Evidence for different relationship variables in the

child and adolescent treatment literature. Clinical Psychology Review, 26, 50–65.

doi:10.1016/j.cpr.2005.09.001

Le Grange, D., Accurso, E. C., Lock, J., Agras, S., & Bryson, S. W. (2014). Early weight gain

predicts outcome in two treatments for adolescent anorexia nervosa. International

Journal of Eating Disorders, 47, 124-129. doi: 10.1002/eat.22221

Leonard, L. M. (2007). A time series design evaluating the effectiveness of a residential

treatment program for eating disorders. Retrieved from Dissertation Abstracts

International: Section B: The Sciences and Engineering. (2007-99200-1280)

Lipsey, M. W., & Wilson, D. B. (2001). Practical meta-analysis. London, UK: Sage

Publications.

Lock, J., Le Grange, D. L., Agras, W. S., & Dare, C. (2001). Treatment manual for anorexia

nervosa: A family-based approach. New York, NY, US: Guilford Press.

Loeb, K. L., Pratt, E. M., Walsh, B.T., Wilson, T.G., Labouvie, E., Hayaki, J., Stewart, A. &

Fairburn, C.G. (2005). Therapeutic alliance and treatment adherence in two interventions

for bulimia nervosa: A study of process and outcome. Journal of Consulting and

Clinical Psychology, 73, 1097–1107. doi: 10.1037/0022-006X.73.6.1097

of 54




Acc

epte

d A

rtic


37

*Mander, J., Teufel, M., Keifenheim, K., Zipfel, S., & Giel, K. E. (2013). Stages of change,

treatment outcome and therapeutic alliance in adult in-patients with chronic anorexia

nervosa. BMC Psychiatry, 13, 111.

Manlick, C. F., Cochran, S. V., & Koon, J. (2013). Acceptance and commitment therapy for

eating disorders: Rationale and literature review. Journal of Contemporary

Psychotherapy, 43, 115-122. doi: 10.1007/s10879-012-9223-7

Margraf, J. (2000). Aus dick wir nicht dünn. Falsche Erwartungshaltungen. In E. Roche (Ed.),

Hülle und Fülle. Dem Fett auf den Leib gerückt Basel, Switzerland: Hoffmann-La Roche.

Martin, S., Sutcliffe, P., Griffiths, F., Sturt, J., Powell, J., Adams, A., & Dale, J. (2011).

Effectiveness and impact of networked communication interventions in young people

with mental health conditions: A systematic review. Patient Education and Counseling,

85, e108-e119. doi:10.1016/j.pec.2010.11.014

McLeod, B. D. (2011). Relation of the alliance with outcomes in youth psychotherapy: A meta-

analysis. Clinical Psychology Review, 31, 603-616.

Miller, K. J., & Mizes, J. S. (2000). Comparative treatments of eating disorders. New

York, NY: Free Association Books.

*Mitchell, J. E., Crosby, R. D., Wonderlich, S. A., Crowe, S., Lancasterb, K., Simonich, H.,…

Myers, T. C. (2008). A randomized trial comparing the efficacy of cognitive–behavioral

therapy for bulimia nervosa delivered via telemedicine versus face-to-face. Behaviour

Research and Therapy, 46, 581–592. doi:10.1016/j.brat.2008.02.004

Morlino, M., Di Pietro, G., Tuccillo, R., Galietta, A., Bolzan, M., Senatore, I., Marozzi, M., &

Valoroso, L. (2007). Drop-out rate in eating disorders: Could it be a function of patient –

therapist relationship? Eating and Weight Disorders, 12, 64-67.

of 54




Acc

epte

d A

rtic


38

Munsch, S., Biedert, E., & Keller, U. (2003). Evaluation of a lifestyle change program for the

treatment of obesity in general practice. Swiss Medical Weekly, 133, 148–154.

*Paulson Karlsson, G. P., Clinton, D., & Nevonen, L. (2013). Prediction of weight increase in

anorexia nervosa. Nordic Journal of Psychiatry, 67, 424-432. doi:

10.3109/08039488.2012.754051

*Prestano, C., Lo Coco, G., Gullo, S., & Lo Verso, G. (2008). Group analytic therapy for eating

disorders: Preliminary results in a single-group study. European Eating Disorders

Review, 16, 302–310. doi: 10.1002/erv.871

Raykos, B. C., Watson, H. J., Fursland, A., Byrne, S. M., & Nathan, P. (2013). Prognostic value

of rapid response to enhanced cognitive behavioral therapy in a routine clinic sample of

eating disorder patients. International Journal of Eating Disorders, 46, 764-770. doi:

10.1002/eat.22169

*Satir, D. (2012). An experimental analysis of alliance focused treatment for anorexia nervosa

(Doctoral dissertation). Retrieved from ProQuest Dissertations and Theses. (Accession

Order No. AAT 3500649)

Shirk, R. S., Karver, M. S., & Brown, R. (2011). The alliance in child and adolescent

psychotherapy. Psychotherapy, 48, 17-24. doi: 10.1037/a0022181

*Simpson, S., Bell, L., Knox, J., & Mitchell, D. (2005). Therapy via videoconferencing: A route

to client empowerment? Clinical Psychology and Psychotherapy, 12, 156–165. doi:

10.1002/cpp.436

*Sly, R., Morgan, J. F., Mountford, V. A., & Lacey, J. H. (2013). Predicting premature

termination of hospitalised treatment for anorexia nervosa: The roles of therapeutic

of 54




Acc

epte

d A

rtic


39

alliance, motivation, and behaviour change. Eating Behaviors, 14, 119–123.

doi.org/10.1016/j.eatbeh.2013.01.007

Sly, R., Morgan, J. F., Mountford, V. A., Sawer, F., Evans, C., & Lacey, J. H. (2014). Rules of

engagement: Qualitative experiences of therapeutic alliance when receiving in-patient

treatment for anorexia nervosa. Eating Disorders: The Journal of Treatment &

Prevention, 22, 233-243. doi.org/10.1080/10640266.2013.867742

Sperry, S., Roehrig, M., & Thompson, J. K. (2009). Treatment of eating disorders in childhood

and adolescence. In L. Smolak & J. K. Thompson (Eds.), Body image, eating

disorders, and obesity in youth: Assessment, prevention, and treatment. (2nd ed.) (pp.

261-279). Washington, DC: American Psychological Association.

*Stiles-Shields, C., Touyz, S., Hay, P., Lacey, H., Crosby, R. D., Rieger, E., … Le Grange, D.

(2013). Therapeutic alliance in two treatments for adults with severe and enduring

anorexia nervosa. International Journal of Eating Disorders, 46, 783–789. doi:

10.1002/eat.22187

Strober, M. (2004). Managing the chronic, treatment- resistant patient with anorexia nervosa.

International Journal of Eating Disorders, 36, 245–255. doi: 10.1002/eat.20054

Tang, T. Z. & DeRubeis, R. J. (1999). Sudden gains and critical sessions in cognitive

behavioural therapy for depression. Journal of Consulting and Clinical Psychology, 67,

894-904.

*Tasca, G. A., Balfour, L., Ritchie, K., & Bissada, H. (2007). The relationship between

attachment scales and group alliance growth differs by treatment type for women with

binge-eating disorder. Group Dynamics: Theory, Research, and Practice, 11, 1–14. doi:

10.1037/1089-2699.11.1.1

of 54




Acc

epte

d A

rtic


40

*Tasca, G. A. & Lampard, A. M. (2012). Reciprocal influence of alliance to the group and

outcome in day treatment for eating disorders. Journal of Counseling Psychology, 59,

507–517. doi: 10.1037/a0029947

*Tasca, G. A., Ritchie, K., Demidenko, N., Balfour, L., Krysanski, V., Weekes, K., …Bissada,

H. (2013). Matching women with binge eating disorder to group treatment based on

attachment anxiety: Outcomes and moderating effects. Psychotherapy Research, 23, 301-

314. doi.org/10.1080/10503307.2012.717309

*Thompson-Brenner, H., Shingleton, R. M., Sauer-Zavala, S., Richards, L. K., & Pratt, E. M.

(2015). Multiple measures of rapid response as predictors of remission in cognitive

behavior therapy for bulimia nervosa. Behaviour Research and Therapy, 64, 9-14.

doi.org/10.1016/j.brat.2014.11.004

Thomas, J. J., Vartanian, L. R., & Brownell, K. D. (2009). The relationship between eating

disorder not otherwise specified (EDNOS) and officially recognized eating disorders:

meta-analysis and implications for DSM. Psychological Bulletin, 135(3), 407. doi:

10.1037/a0015326

Treasure, J. L., Katzman, M., Schmidt, U., Troop, N., Todd, G., & de Silva, P. (1999).

Engagement and outcome in the treatment of bulimia nervosa: First phase of a sequential

design comparing motivation enhancement therapy and cognitive behavioural therapy.

Behavioural Research and Therapy, 37, 405–418. doi:10.1016/S0005-7967(98)00149-1

Vitousek, K. & Watson, S. (1998). Enhancing motivation for change in treatment-resistant eating

disorders. Clinical Psychology Review, 18, 391-420. doi:10.1016/S0272-7358(98)00012-

9

of 54




Acc

epte

d A

rtic


41

Vocks, S., Tuschen-Caffier, B., Pietrowsky, R., Rustenbach, S. J., Kersting, A., & Herpertz, S.

(2010). Meta-analysis of the effectiveness of psychological and pharmacological

treatments for binge eating disorder. International Journal of Eating Disorders, 43, 205-

217. doi: 10.1002/eat.20696

*Waller, G., Evans, J., & Stringer, H. (2012). The therapeutic alliance in the early part of

cognitive behavioral therapy for the eating disorders. International Journal of Eating

Disorders, 45, 63–69. doi: 10.1002/eat.20914

Westwood, L. M. & Kendal, S. E. (2012). Adolescent client views towards the treatment of

anorexia nervosa: A review of the literature. Journal of Psychiatry and Mental Health

Nursing, 19, 500-508. doi: 10.1111/j.1365-2850.2011.01819.x

Wilson, G. T. (2011). Treatment of binge eating disorder. Psychiatric Clinics of North America,

34:773–783.

Wilson, G. T., Loeb, K. L., Walsh, B. T., L, E., Petkova, E., Liu, X., & Waternaux, C. (1999).

Psychological versus pharmacological treatments of bulimia nervosa: Predictors and

processes of change. Journal of Consulting and Clinical Psychology, 67(4), 451-459. doi:

10.1037/0022-006X.67.4.451

*Zaitsoff, S. L., Doyle, A. C., Hoste, R.R., & Le Grange, D. (2008). How do adolescents with

bulimia nervosa rate the acceptability and therapeutic relationship in family-based

treatment? International Journal of Eating Disorders, 41, 390–398. DOI:

10.1002/eat.20515

Zeeck, A. & Hartmann, A. (2005). Relating therapeutic process to outcome: Are there predictors

for the short-term course in anorexic patients? European Eating Disorders Review, 13,

245–254. doi: 10.1002/erv.646

of 54




Acc

epte

d A

rtic


42

of 54




Acc

epte

d A

rtic

leMETA-ANALYSIS OF THE RELATION BETWEEN ALLIANCE

Figure 2. Forest plots for all meta-analytic research questions.

of 54




Acc

epte

d A

rtic


Table 1. Question 1: Does early symptom change predict early/mid all?

Moderator variables Effect size information

Report N Mean

age DX

Therapy

setting

Therapy

type/mode

Alliance

rater

Session/

of Total

Study

drop-

out (%)

Alliance

rating

measure

β(SE) 95% CI p

Bourion-Bedes et al. (2013)(A) 66 15.30 AN Inpatient Multi/IND PA 3/VAR 0 HAQ -.01(.13)b -.26, .23 .92

Bourion-Bedes et al. (2013)(B) 42 15.30 AN Outpatient Multi/IND PA 3/VAR 0 HAQ .19(.17)b -.15, .53 .28

Brown et al. (2013) 35 25.70 AN Outpatient CBT/IND PA 6/30-40 32.31 WAI .46(.19)c .10, .83 .01

Constantino et al. (2005)(A) 75 28.10 BN Outpatient CBT/IND PA 12/19 25.91 HAQ .54(.19)c .17, .90 < .01

Constantino et al. (2005)(B) 82 28.10 BN Outpatient IPT/IND PA 12/19 25.91 HAQ .40(.12)c .16, .63 < .01

Forsberg et al. (2013)(A) 40 14.80 AN Outpatient AFT/IND IO 3-5/32 17.36 WAI -.05(.17)b -.38, .29 .78

Forsberg et al. (2013)(B) 38 14.00 AN Outpatient FBT/IND IO 3-5/20 17.36 WAI .35(.17)b .02, .67 .04

Isserlin & Couturier (2012) 13 14.00 AN Outpatient FBT/IND IO 3/MDN=12 42.86 SOFTA .55(.39) -.22, 1.32 .16

Paulson Karlsson et al. (2013) 41 23.90 AN MIX Multi/MIX PA MO6/

MO18±19 38.00 TSS .26(.15)c -.02, .55 .07

Prestano et al. (2008) 6 16.00 MUL Outpatient Other/GRP PA WK4/WK104

25.00 CAPAS .07(.81)ab

-1.52,

1.65 .93

Satir (2012) 6 26.90 AN Outpatient Multi/IND PA 5/24 14.29 WAI -.20(.50)b -1.17, .78 .69

Simpson et al. (2005) 6 32.00 BN Outpatient CBT/IND PA 4/17

0 ARM -.38(.77)b -1.88,

1.13 .63

Sly et al. (2013) 78 27.73 AN Inpatient Other/IND PA WK4/VAR 0 WAI .23(.11)c .02, .45 .03

Tasca & Lampard (2012) 127 26.11 MUL Outpatient Multi/GRP PA WK4/WK12 28.00 CAPAS .16(.08)c -.01, .32 .06

Thompson-Brenner (2013) 36 25.63 BN Outpatient CBT/IND PA 3-5/20

24.00 WAI -.52(1.02)c -2.52,

1.48 .61

Waller et al.(2012) 44 27.20 MUL Outpatient CBT/IND PA 6/6 14.00 WAI .18(.02)c .15, .21 < .01

Zaitsoff et al.(2008)(A) 33 11.25 BN Outpatient FBT/IND PA 10/20 11.25 HRQ -.39(.18)ac -.74, -.04 .03

Zaitsoff et al.(2008)(B) 29 11.25 BN Outpatient SPT/IND PA 10/20 11.25 HRQ .35(.19)ac -.02, .72 .07

of 54




Acc

epte

d A

rtic


Note. β = .19, 95% CI [.11, .28], z = 4.38, p < .0001; A = treatment arm A; B = treatment arm B; DX = sample diagnosis; AN = anorexia nervosa;

BN = bulimia nervosa; MUL = multiple eating disorder diagnoses; MIX = multiple settings; CBT = cognitive behavioral therapy; IPT =

interpersonal psychotherapy; AFT = adolescent-focused therapy; FBT = family-based therapy; SPT = supportive psychotherapy; IND = individual;

GRP = group; PA = patient; IO = independent observer; VAR = varied; MDN = median; MO = months; WK = week; HAQ = Helping Alliance

Questionnaire; WAI = Working Alliance Inventory; SOFTA = System for Observing Family Alliances; TSS = Treatment Satisfaction Scale;

CAPAS = The California Psychotherapy Alliance Scales; ARM = Agnew Relationship Measure; HRQ = Helping Relationship Questionnaire; CI =

confidence interval. aThe baseline measure of the outcome was not included in the regression analysis due to multicolinearity.

bThe study effect size was based on total sample analyses.

cThe study effect size was based on pairwise regression analyses.

of 54




Acc

epte

d A

rtic


Table 2. Results of random-effects moderator analyses

Does early change in

treatment outcome

predict early/mid

alliance?

Does early/mid

alliance predict

subsequent symptom

change?

Does early/mid

alliance predict

subsequent symptom

change above and

beyond early

symptoms change?

Moderator variable Q df P Q df p Q df p

Therapy type 1.56 3 .67 10.61 4 .03 5.89 3 .12

Mean age 2.77 1 .10 1.03 1 .31 16.20 1 < .01

Eating disorder diagnosis .07 2 .97 .60 3 .90 6.10 2 .047

Alliance rater .01 1 .93 .25 1 .62 1.53 1 .22

Study drop-out rate 3.65 1 .06 .63 1 .43 .95 1 .33

of 54




Acc

epte

d A

rtic


Table 3. Question 2: Does mid-to-end of treatment change in symptoms predict later alliance?


Report N Mean

age DX

Therapy

setting

Therapy

type/mode

Alliance

rater/session

Study

drop-out

(%)

Alliance

rating

measure

β(SE) 95% CI p

Brown et al. (2013) 33 25.70 AN Outpatient CBT/IND PA/EOT 32.31 WAI .03(.21)b -.39, .44 .90

Fluckiger et al. (2011)(A) 29 45.93 BED Outpatient CBT/IND PA/EOT 29.00 BPSRP -.06(.20)a -.44, .33 .77

Fluckiger et al. (2011)(B) 26 45.93 BED Outpatient BWLT/IND PA/EOT 29.00 BPSRP .07(.21)a -.33, .48 .73

Isserlin & Couturier (2012) 14 14.00 AN Outpatient FBT/IND IO/EOT 42.86 SOFTA .26(.22)b -.17, .69 .23

Prestano et al. (2008) 6 16.00 MUL Outpatient Other/GRP PA/EOT 25.16 CAPAS .37(.49)a -.59, 1.33 .45

Simpson et al. (2005) 6 32.00 BN Outpatient CBT/IND PA/EOT 0 ARM .41(.60)a -.77, 1.59 .50

Tasca & Lampard (2012) 65 26.11 MUL Outpatient Multi/GRP PA/EOT 28.00 CAPAS .17(.13)b -.10, .43 .22

Tasca et al. (2013) 49 44.30 BED Outpatient IPT/GRP PA/EOT 18.00 CAPAS .12(.19)b -.24, .48 .52

Zaitsoff et al.(2008)(A) 28 16.10 BN Outpatient FBT/IND PA/EOT 11.25 HRQ .02(.21)b -.40, .43 .93

Zaitsoff et al.(2008)(B) 24 16.10 BN Outpatient SPT/IND PA/EOT 11.25 HRQ -.26(.52)b -1.28, .76 .62

Note. β = .10, 95% CI [-.04, .24], z = 1.46, p = .15.; A = treatment arm A; B = treatment arm B; DX = sample diagnosis; AN = anorexia

nervosa; BN = bulimia nervosa; BED = binge eating disorder; MUL = multiple eating disorder diagnoses; CBT = cognitive behavioral

therapy; IPT = interpersonal psychotherapy; BWLT = behavioral weight loss treatment; AFT = adolescent-focused therapy; FBT = family-

based therapy; SPT = supportive psychotherapy; IND = individual; GRP = group; PA = patient; IO = independent observer; EOT = end of

treatment; WAI=Working Alliance Inventory; BPSRP=Bern Post-Session Reports for Patients; SOFTA = System for Observing Family

Therapy Alliances; CAPAS = The California Psychotherapy Alliance Scales; ARM = Agnew Relationship Measure; HRQ = Helping

Relationship Questionnaire; CI = confidence interval. aThe study effect size was based on total sample analyses.

bThe study effect size was based on pairwise regression analyses.

of 54




Acc

epte

d A

rtic


Table 4. Question 3: Does change in symptoms across treatment predict later alliance?


Report N Mean

age DX

Therapy

setting

Therapy

type/mode

Alliance

rater/

session

Study

drop-out

(%)

Alliance

rating

measure

β(SE) 95% CI p

Brown et al. (2013) 31 25.70 AN Outpatient CBT/IND PA/EOT 32.31 WAI .11(.21)c -.29, .51 .58

Fluckiger et al. (2011)(A) 29 45.93 BED Outpatient CBT/IND PA/EOT 29.00 BPSRP .17(.20)b -.21, .56 .37

Fluckiger et al. (2011)(B) 26 45.93 BED Outpatient BWLT/IND PA/EOT 29.00 BPSRP .17(.21)b -.23, .58 .41

Isserlin & Couturier (2012) 14 14.00 AN Outpatient FBT/IND IO/EOT 42.86 SOFTA .38(.26)c -.12, .89 .14

Mander et al. (2013) 39 27.70 AN Inpatient Multi/MIX PA/EOT 28.00 SACiP .21(.17)b -.12, .54 .21

Mitchell et al. (2008)(A) 35 29.60 BN Outpatient CBT/IND PA/EOT 37.50 WAI .57(.29)c .01, 1.13 .05

Mitchell et al. (2008)(B) 36 28.40 BN Outpatient CBT/IND PA/EOT 37.50 WAI .09(.45)ac -.80, .98 .85

Prestano et al. (2008) 6 16.00 MUL Outpatient Other/GRP PA/EOT 25.00 CAPAS .42(.66)ab -.87, 1.71 .52

Simpson et al. (2005) 6 32.00 BN Outpatient CBT/IND PA/EOT 0 ARM .06(.55)b -1.01, 1.13 .91

Stiles-Shields et al. (2013)(A) 24 33.40 AN Outpatient CBT/IND PA/EOT 22.58 HAQ .66(.19)c .29, 1.04 < .01

Stiles-Shields et al. (2013)(B) 28 33.40 AN Outpatient SSCT/IND PA/EOT 12.50 HAQ .33(.20)c -.06, .72 .10

Tasca & Lampard (2012) 65 26.11 MUL Outpatient Multi/GRP PA/EOT 28.00 CAPAS .20(.13)c -.06, .46 .13

Tasca et al.(2007)(A) 38 43.86 BED Outpatient CBT/GRP PA/EOT 22.73 CAPAS .08(.20)c -.30, .47 .68

Tasca et al.(2007)(B) 52 43.86 BED Outpatient IPT/GRP PA/EOT 22.73 CAPAS -.20(.15)c -.50, .10 .19

Tasca et al. (2013) 72 44.30 BED Outpatient IPT/GRP PA/WK16 18.00 CAPAS .12(.13)c -.14, .39 .35

Thompson-Brenner et al. (2013) 37 25.63 BN Outpatient CBT/IND PA/14-EOT 24.00 WAI .42(.24)c -.05, .89 .08

Zaitsoff et al.(2008)(A) 29 16.10 BN Outpatient FBT/IND PA/EOT 11.25 HRQ .14(.20)ac -.24, .52 .49

Zaitsoff et al.(2008)(B) 31 16.10 BN Outpatient SPT/IND PA/EOT 11.25 HRQ -.32(.19)ac -.69, .06 .10

Note. β = .17, 95% CI [.06, .29], z = 2.96, p = .003; A = treatment arm A; B = treatment arm B; DX = sample diagnosis; AN = anorexia

nervosa; BN = bulimia nervosa; BED = binge eating disorder; MUL = multiple eating disorder diagnoses; CBT = cognitive behavioral

therapy; IPT = interpersonal psychotherapy; BWLT = behavioral weight loss treatment; AFT = adolescent-focused therapy; FBT = family-

based therapy; SPT = supportive psychotherapy; SSCT= specialist supportive clinical management; IND = individual; GRP = group; PA =

patient; IO = independent observer; WK = week; EOT = end of treatment; WAI=Working Alliance Inventory; BPSRP=Bern Post-

Session Reports for Patients; SOFTA = System for Observing Family Therapy Alliances; SACiP = Scale for the Multiperspective

Assessment of General Change Mechanisms in Psychotherapy; CAPAS = The California Psychotherapy Alliance Scales; ARM = Agnew

Relationship Measure; HAQ = Helping Alliance Questionnaire; HRQ = Helping Relationship Questionnaire; CI = confidence interval. aThe baseline measure of the outcome was not included in the regression analysis due to multicolinearity.

of 54




Acc

epte

d A

rtic


bThe study effect size was based on total sample analyses.

cThe study effect size was based on pairwise regression analyses.

of 54




Acc

epte

d A

rtic


Table 5. Question 4a: Does early/mid alliance predict subsequent change in symptoms?


Report N Mean

age DX

Therapy

setting

Therapy

type/mode

Alliance

rater

Session/

of Total

Study

drop-

out

(%)

Alliance

rating

measure β(SE) 95% CI p

Bourion-Bedes et al. (2013)(A) 66 15.30 AN Inpatient Multi/IND PA 3/VAR 0 HAQ .27(.12)a .04, .51 .02

Bourion-Bedes et al. (2013)(B) 42 15.30 AN Outpatient Multi/IND PA 3/VAR 0 HAQ .35(.16)a .04, .66 .03

Brown et al. (2013) 33 25.70 AN Outpatient CBT/IND PA 6/30-40 32.31 WAI -.25(.15)b -.54, .05 .10

Constantino et al. (2005)(A) 72 28.10 BN Outpatient CBT/IND PA 12/19 25.91 HAQ .11(.11)b -.10, .32 .31

Constantino et al. (2005)(B) 76 28.10 BN Outpatient IPT/IND PA 12/19 25.91 HAQ -.05(.08)b -.21, .12 .56

Fluckiger et al. (2011)(A) 29 45.93 BED Outpatient CBT/IND PA 6/22 29.00 BPSRP .13(.16)a -.25, .51 .51

Fluckiger et al. (2011)(B) 26 45.93 BED Outpatient BWLT/IND PA 6/22 29.00 BPSRP -.03(.22)a -.45, .40 .90

Forsberg et al. (2013)(A) 40 14.80 AN Outpatient AFT/IND IO 3-5/20 17.36 WAI .13(.16)a -.18, .44 .41

Forsberg et al. (2013)(B) 38 14.00 AN Outpatient FBT/IND IO 3-5/20 17.36 WAI .23(.16)a -.09, .55 .16

Isserlin & Couturier (2012) 13 14.00 AN Outpatient FBT/IND IO 3/

MDN=12 42.86

SOFTA .25(.33)b -.40, .90 .46


MO18±19 38.00

TSS .21(.15)b -.09, .50 .18

Mander et al. (2013) 39 27.70 AN Inpatient Multi/MIX PA DAY1/M=

DAY48.8 28.00

SACiP .37(.17)a .04, .70 .03

Prestano et al. (2008) 6 16.00 MUL Outpatient Other/GRP PA WK4/

WK104 25.00

CAPAS .59(.77)a

-.93,

2.01 .45

Simpson et al. (2005) 6 32.00 BN Outpatient CBT/IND PA 4/17

0 ARM

-.56(.60)a -1.74,

.62 .35

Sly et al. (2013) 78 27.73 AN Inpatient Other/IND PA WK4/

VAR 0

WAI -.10(.13)b -.35, .15 .44

Tasca & Lampard (2012) 89 26.11 MUL Outpatient Multi/GRP PA WK4/

WK12 28.00

CAPAS .13(.09)b -.04, .30 .14

Tasca et al. (2013) 50 44.30 BED Outpatient IPT/GRP PA WK4/

WK16 18.00

CAPAS .26(.11)b .05, .48 .02

Zaitsoff et al.(2008)(A) 28 16.10 BN Outpatient FBT/IND PA 10/20 11.25 HRQ .45(.20)b .06, .84 .03

Zaitsoff et al.(2008)(B) 26 16.10 BN Outpatient SPT/IND PA 10/20 11.25 HRQ .14(.20)b -.26, .54 .50

of 54




Acc

epte

d A

rtic


Note. β = .13, 95% CI [.05, .22], z = 3.10, p = .002; A = treatment arm A; B = treatment arm B; DX = sample diagnosis; AN = anorexia nervosa;

BN = bulimia nervosa; BED = binge eating disorder; MUL = multiple eating disorder diagnoses; CBT = cognitive behavioral therapy; IPT =

interpersonal psychotherapy; BWLT = behavioral weight loss treatment; AFT = adolescent-focused therapy; FBT = family-based therapy; SPT =

supportive psychotherapy; IND = individual; GRP = group; PA = patient; IO = independent observer; VAR = varied; MDN = median; MO = month;

DAYS = days; WK = week; HAQ = Helping Alliance Questionnaire; WAI=Working Alliance Inventory; BPSRP = Bern Post-Session Reports for

Patients; SOFTA = System for Observing Family Therapy Alliances; TSS = Treatment Satisfaction Scale; SACiP = Scale for the Multiperspective

Assessment of General Change Mechanisms in Psychotherapy; CAPAS = The California Psychotherapy Alliance Scales; ARM = Agnew

Relationship Measure; HRQ = Helping Relationship Questionnaire; CI = confidence interval. aThe study effect size was based on total sample analyses.


of 54




Acc

epte

d A

rtic


Table 6. Question 4b: Does early/mid alliance predict subsequent change in symptoms above and beyond early change in symptoms?


Report N Mean

age DX

Therapy

setting

Therapy

type/mode

Alliance

rater

Session/

of Total

Study

drop-

out (%)

Alliance

rating

measure

β(SE) 95% CI p

Bourion-Bedes et al. (2013)(A) 66 15.30 AN Inpatient Multi/IND PA 3/VAR 0 HAQ .31(.11)a .09, .52 .01

Bourion-Bedes et al. (2013)(B) 42 15.30 AN Outpatient Multi/IND PA 3/VAR 0 HAQ .27(.14)a -.01, .55 .05

Brown et al. (2013) 33 25.70 AN Outpatient CBT/IND PA 6/30-40 32.31 WAI -.14(.19)b -.52, .23 .46

Constantino et al. (2005)(A) 72 28.10 BN Outpatient CBT/IND PA 12/19 25.91 HAQ -.15(.12)b -.39, .09 .21

Constantino et al. (2005)(B) 76 28.10 BN Outpatient IPT/IND PA 12/19 25.91 HAQ -.20(.12)b -.43, .03 .09

Forsberg et al. (2013)(A) 40 14.80 AN Outpatient AFT/IND IO 3-5/32 17.36 WAI .17(.16)a -.14, .47 .28

Forsberg et al. (2013)(B) 38 14.00 AN Outpatient FBT/IND IO 3-5/20 17.36 WAI .25(.18)a -.10, .60 .16

Isserlin & Couturier (2012) 13 14.00 AN Outpatient FBT/IND IO 3/

MDN=12 42.86

SOFTA .30(.34)b -.36, .96 .37


MO18±19 38.00

TSS .19(.15)b -.11, .50 .21

Prestano et al. (2008) 6 16.00 MUL Outpatient Other/GRP PA WK4/

WK104 25.00

CAPAS -.07(.77)a -.82, .69 .86

Simpson et al. (2005) 6 32.00 BN Outpatient CBT/IND PA 4/17 0 ARM -.35(.42)a -1.18, .48 .41

Sly et al. (2013) 78 27.73 AN Inpatient Other/IND PA 4/VAR 0 WAI -.07(.12)b -.31, .16 .53

Tasca & Lampard (2012) 89 26.11 MUL Outpatient Multi/GRP PA WK4/

WK12 28.00

CAPAS .07(.11)b -.13, .28 .48

Zaitsoff et al.(2008)(A) 28 16.10 BN Outpatient FBT/IND PA 10/20 11.25 HRQ -.03(.21)b -.43, .37 .88

Zaitsoff et al.(2008)(B) 26 16.10 BN Outpatient SPT/IND PA 10/20 11.25 HRQ .28(.20)b -.11, .68 .16

Note. β = .07, 95% CI [-.04, .17], z = 1.26, p = .21; A = treatment arm A; B = treatment arm B; DX = sample diagnosis; AN = anorexia nervosa;

BN = bulimia nervosa; BED = binge eating disorder; MUL = multiple eating disorder diagnoses; CBT = cognitive behavioral therapy; IPT =

interpersonal psychotherapy; BWLT = behavioral weight loss treatment; AFT = adolescent-focused therapy; FBT = family-based therapy; SPT =

supportive psychotherapy; IND = individual; GRP = group; PA = patient; IO = independent observer; VAR = varied; MDN = median; MO =

month; WK = week; HAQ = Helping Alliance Questionnaire; WAI=Working Alliance Inventory; HAQ = Helping Alliance Questionnaire;

SOFTA = System for Observing Family Therapy Alliances; TSS = Treatment Satisfaction Scale; CAPAS = The California Psychotherapy

Alliance Scales; ARM = Agnew Relationship Measure; HRQ = Helping Relationship Questionnaire; CI = confidence interval. aThe study effect size was based on total sample analyses.


of 54




Acc

epte

d A

rtic


Acknowledgements

This meta-analysis was truly a multi-site study. Most of the included papers did not report all of the

necessary information in the original published reports to enable our Boston-based research team to

answer each meta-analytic research question. For example, some papers reported alliance at the

beginning of treatment and its relation to symptom change at the end of treatment, but not to

symptom change early in treatment. Therefore, to make this study possible, we reached out to each

of the authors of the papers that met inclusion criteria. We asked if these authors could provide their

raw data, so that we could re-calculate effect sizes for each study and combine them together for the

current meta-analysis. To acknowledge the important contribution of these raw data, we invited the

first (or, in some cases, corresponding) author on each of the included studies to be a co-author on

the current meta-analysis. For 5 of the 20 studies, we included more than one co-author, in

recognition that both co-authors assisted in preparing raw data for inclusion.

of 54




Acc

epte

d A

rtic

leIJED-16-0463.R2

Meta-Análisis de la Relación entre la Alianza Terapéutica y el Resultado del Tratamiento

en los Trastornos de la Conducta Alimentaria.

Resumen: Objetivo: La alianza terapéutica entre paciente y terapeuta ha demostrado ser una

relación con resultados psicoterapéuticos favorables en el tratamiento de los trastornos de la

conducta alimentaria (TCA). Sin embargo, quedan preguntas acerca de la inter-relación entre

alianza temprana, mejoría temprana de síntomas y resultados del tratamiento. Hicimos un meta-

análisis de la relación entre estos constructos y los posibles moderadores de estas relaciones en

los tratamientos psicosociales para TCA. Método: Veintiún estudios reunieron los criterios de

inclusión y aportaron suficientes datos suplementarios. Resultados: los resultados revelaron un

efecto de la talla pequeño a moderado, ฆยs = .13 a .22 (p < .05), encontrando que la mejoría

temprana de los síntomas estuvo relacionada con la subsecuente calidad de la alianza y las

calificaciones de la alianza también estuvieron relacionadas con la subsecuente reducción de los

síntomas. La relación entre alianza temprana y resultados de tratamiento fue parcialmente

explicada por la temprana mejoría de los síntomas. Con relación a los moderadores, la alianza

temprana mostró débiles asociaciones con el resultado en terapias con un fuerte componente

conductual relativo a terapias no conductuales. Sin embargo, la alianza mostró más fuerte

relación con los resultados para pacientes más jóvenes (versus mayores), por encima y sobre la

varianza compartida con la temprana mejoría de síntomas. Discusión: En resumen, la reducción

temprana de los síntomas refuerza la alianza terapéutica y los resultados del tratamiento en TCA,

pero la alianza temprana puede requerir atención específica para pacientes jóvenes y para

aquellos que no reciben tratamientos basados en una orientación conductual.

of 54




thomas, jennifer; massachusetts general hospital, eating

Documents